Simple tricks for improving pattern-based information extraction from the biomedical literature

Abstract

BackgroundPattern-based approaches to relation extraction have shown very good results in many areas of biomedical text mining. However, defining the right set of patterns is difficult; approaches are either manual, incurring high cost, or automatic, often resulting in large sets of noisy patterns.ResultsWe propose several techniques for filtering sets of automatically generated patterns and analyze their effectiveness for different extraction tasks, as defined in the recent BioNLP 2009 shared task. We focus on simple methods that only take into account the complexity of the pattern and the complexity of the texts the patterns are applied to. We show that our techniques, despite their simplicity, yield large improvements in all tasks we analyzed. For instance, they raise the F-score for the task of extraction gene expression events from 24.8% to 51.9%.ConclusionsAlready very simple filtering techniques may improve the F-score of an information extraction method based on automatically generated patterns significantly. Furthermore, the application of such methods yields a considerable speed-up, as fewer matches need to be analysed. Due to their simplicity, the proposed filtering techniques also should be applicable to other methods using linguistic patterns for information extraction.

An up-to-date knowledge-based literature search and exploration framework for focused bioscience domains

“…Most of these systems target news articles or Web pages for relations such as "Company located at Headquarters", "Player plays for Team", or "Person born in City". In the biomedical area, pattern based approaches have been mostly used to extract very specific types of relationships such as protein-protein interactions [15]. Our current work [21] is inspired by Turney's [22] vector space methods and is based on frequency distributions of automatically extracted patterns.…”

Extraction de relations complexes. Application à des résultats expérimentaux en physiologie rénale

“…» Trois sous-tâches étaient proposées : a) la détection du déclencheur de l'événe-ment ainsi que son typage, et la reconnaissance de la protéine (ou le gène) qui subit ce changement d'état, b) la reconnaissance du deuxième argument, et c) la détection d'un modifieur de l'événement, par exemple si l'événement est à la forme négative dans le texte. (Björne et al, 2010) associent de l'apprentissage supervisé et l'utilisation de règles pour la détection des événements, et (Nguyen et al, 2010) font de l'apprentissage de patrons. (Buyko et al, 2009), qui sont arrivés deuxièmes pour la tâche 1 de BioNLP'09 avec une F-mesure 3 de 0,46 %, décrivent la tâche d'extraction d'événe-ment en six sous-tâches (entre parenthèses, nous indiquons le résultat de l'extraction de l'événement de l'exemple donné précédemment) :…”

ANDDigest: a new web-based module of ANDSystem for the search of knowledge in the scientific literature

“…Unlike the co-occurrence-based methods, the use of manually defined rules and templates often allows authors to achieve high values of accuracy, but they tend to have low completeness [26]. Several automated methods of rules and templates generation were proposed for dealing with this problem [27,28]. RLIMS-P [29,30] and MinePhos [31] are examples of template-based tools, both using rule-based templates for mining information on phosphorylation from the literature.…”